GPU Ray Tracing Performance Comparisons [2021-2022]

Battlefield 2042 seems to scale well with core count (at least up to 8 cores ... could be partially why a 3950 underperforms with 16 cores)

But ray tracing just has a serious hit, and ambient occlusion isn't worth it. You don't really notice it anywhere, though the guy in this vid seems to spend all of his time looking at puddles lol

Looks like there was a nice improvement from the beta in performance, but it's overall not high enough to make RT hit worth it.
 
Last edited:
My cpu in battlefield 2042 .. min/avg/max ... wonder why ray tracing kills me. This is without ray tracing, lol.

View attachment 6012
So basically an 8-threaded game on average? Shame that no tool (that I know of) can take a max with ind. threads for the maximum overall load at a given point in time. For all we know from normal tools, those spikes could be very well distributed temporally.
 
This CPU load actually looks okay.
It was always bound to happen that CPU load would increase considerably with the launch of new console gen.
7 Zen2 cores at 3.2-3.6 GHz will likely be maxed out during this generation even in 30 fps titles.
So to get above the same console fps target on PC we'd need considerably faster CPUs.
 
This CPU load actually looks okay.
It was always bound to happen that CPU load would increase considerably with the launch of new console gen.
7 Zen2 cores at 3.2-3.6 GHz will likely be maxed out during this generation even in 30 fps titles.
So to get above the same console fps target on PC we'd need considerably faster CPUs.
I’m really looking forward to what devs put all that CPU power towards in 30 fps console games.
 
At the same time bf2042 is still a cross gen game and will be out on PS4 too. Granted limited to 64 players, but still...
 
Worst-case on this overclocked 3090 is about 2.8ms for ray tracing (e.g. 81 vs 66fps or 90 vs 72):


5:24 if you just want to see the performance comparison.

Guessing: the ray-tracing hit varies according to how much async-compute headroom there is. Other ideas?

The performance hit of ray tracing also reduces power consumption slightly.
 
Worst-case on this overclocked 3090 is about 2.8ms for ray tracing (e.g. 81 vs 66fps or 90 vs 72):


5:24 if you just want to see the performance comparison.

Guessing: the ray-tracing hit varies according to how much async-compute headroom there is. Other ideas?

The performance hit of ray tracing also reduces power consumption slightly.
The person here is using the 2 bounce RT, and not the default single bounce. So I imagine it could be the shading cost depending upon the angle of the car? If the camera perspective allows the second bounce to actually hit the car instead of miss and fall back to the cubemap, then there would be the incurred shading cost for that second car hit.
 
Worst-case on this overclocked 3090 is about 2.8ms for ray tracing (e.g. 81 vs 66fps or 90 vs 72):


5:24 if you just want to see the performance comparison.

Guessing: the ray-tracing hit varies according to how much async-compute headroom there is. Other ideas?

The performance hit of ray tracing also reduces power consumption slightly.

Technically interesting. But after seeing that much footage of gameplay, I understand what DF meant when they said it's not noticable during normal gameplay.

So, you basically take a fairly hefty hit to performance for virtually no benefit when playing the game. Although I'm sure it'll make photomode nicer, I can totally see why Playground Games didn't bother to implement it outside of Vista mode.

Still, it's an interesting technical achievement.

Regards,
SB
 
Its impressive theres any ray tracing at all considering the game doesnt even launch with it. Its another advantage of RT, community/modders can add it themselfs.
 
The person here is using the 2 bounce RT, and not the default single bounce. So I imagine it could be the shading cost depending upon the angle of the car? If the camera perspective allows the second bounce to actually hit the car instead of miss and fall back to the cubemap, then there would be the incurred shading cost for that second car hit.
You can see from here:


that the angle of the car makes no difference.

Anyone who is interested and has the game can try to see what happens in two different environments, I suppose: desert versus jungle. And then they can vary the "world detail" settings in each environment. Will need the same weather and time of day.

I would expect to see that overall frame rendering and reflection-cube-map rendering would interact with the ray-tracing hit. Since the ray-tracing hit seems to vary by a factor of 2 or more, I suppose that the variability in the ray tracing hit will be reduced by desert rendering with low world detail (number of objects).

Can reflection-cube-maps be turned off entirely?
 
Driveclub has convincing self reflections. How did they manage that? What prevents other studios from adopting the solution?
 
Pretty sure it was just a good implementation of screen space reflection.
Most likely they didn't even reshade objects in SSR. (Which could be done, like BFV did in RT mode.)

Its appears so.

They were incredibly smart on that game.

https://andrewbolt.artstation.com/projects/Oyo6EJ

tUiWUV3.png


83O5I19.png


7cpTM1W.png


Strange we needed to wait 7 years and for raytracing to happen to see this again.
 
Back
Top